Bug report: OpenMP in Abinit 6.10.2
Posted: Thu Feb 09, 2012 11:12 am
Hi all,
There are a few issues with OpenMP use in Abinit 6.10.2 (I haven't checked 6.12.1 yet - hard to keep up with the releases!). It was difficult to figure out what was going on until I switched to ifort (12.1) which has better OpenMP error reporting. The issues I've found so far are:
1. In sphere_fft.F90, the two regions of code starting at lines 102 and 232 have OMP directives enclosing commented-out code - the OMP directives need a further comment character in order to deactivate them otherwise they make the compiler think there is an empty parallel region.
2. In back.F90 (52_fft_mpi_noabirule), there are examples of invalid variable lists on OMP directives - the current generation of compilers (at least GCC 4.6, Intel 12.1) don't accept subobjects like my_type%some_variable in OMP directives as far as I can tell.
I tried to fill out a bug report form as per the "how to help the Abinit Developers" page but the link is broken...
Regards,
Kane
There are a few issues with OpenMP use in Abinit 6.10.2 (I haven't checked 6.12.1 yet - hard to keep up with the releases!). It was difficult to figure out what was going on until I switched to ifort (12.1) which has better OpenMP error reporting. The issues I've found so far are:
1. In sphere_fft.F90, the two regions of code starting at lines 102 and 232 have OMP directives enclosing commented-out code - the OMP directives need a further comment character in order to deactivate them otherwise they make the compiler think there is an empty parallel region.
2. In back.F90 (52_fft_mpi_noabirule), there are examples of invalid variable lists on OMP directives - the current generation of compilers (at least GCC 4.6, Intel 12.1) don't accept subobjects like my_type%some_variable in OMP directives as far as I can tell.
I tried to fill out a bug report form as per the "how to help the Abinit Developers" page but the link is broken...
Regards,
Kane